IEE 572

DESIGN OF ENGINEERING EXPERIMENTS

Specimen Geometry and Aggregate Size

Study for Permanent Deformation in Asphalt
Concrete Pavements

FINAL PROJECT REPORT

Javed Bari
Jong-Suk Jung
Yow-Shi Ho

Date: 12/04/00

Foreword
As part of the course IEE572, this group was looking forward to designing an
engineering experiment. Luckily, the members of this group were already engaged in an
experiment in the Advanced Asphalt Laboratory of the Department of Civil and
Environmental Engineering of Arizona State University. So the group members decided to
design that experiment. In that time, only the first quarter of the course content of the IEE572
course was covered. With that background knowledge, it was not possible to make a
comprehensive design plan. So the group decided to design a factor screening experiment.
As each test took a considerably long time, it was not possible to perform complementary
detailed experiments during the time period of this semester. All the three group-members
actively participated in the design and experiment works.

Introduction
Background of the Experiment
In pavement engineering, permanent deformation (p) is an important parameter
regarding design. Many researchers have earlier observed that test result of p varies with the
size and geometry of test specimens of uniaxial load test using universal testing machine. For
composite materials like asphalt concrete, the ideal test specimen must be large enough
relative to the size of the individual aggregate particles. But the problem is that different
researchers have been using test specimens of different size and geometry for test purpose.
This leads to different sets of database often not compatible to each other. So it becomes hard
to compare one result to another and even harder to use them as basis for further analysis.
These inconsistencies are basically due to the absence of an available and universally agreed
optimum specimen size and geometry. In fact few researches has so far been conducted for
obtaining an optimum specimen size and geometry. As part of their graduate research, the
group members had already been involved in a similar research, though in an early stage. So
the group decided to design and conduct a factor screening experiment with a view to finding
out an optimum specimen size and geometry for testing permanent deformation of asphalt
pavement.

The Experiment

Objectives
The objective of the experiment is to determine the important and optimum factors,
along with their correlation, for a test specimen to be tested in permanent deformation test of
asphalt pavements, which should lead to fairly accurate and consistent permanent deformation
values.

Test Method
The experiment is a uniaxial load test that was conducted with universal testing
machine in the in the Advanced Asphalt Laboratory of the Department of Civil and
Environmental Engineering of Arizona State University. The test specimens were fabricated
in the Superpave Gyratory Compactor. The loading type is repetitive and uniaxial. The
preparation of samples, testing of specimens and collection of data were consistent with the
standard procedures.

Experimental Plan
Hypothesis
The experimental hypotheses are:
1) Minimum specimen dimension has a significant effect on permanent deformation
measured in the uniaxial test.
2) There is a limiting minimum specimen dimension above which the material properties are
independent of this dimension. This dimension is the optimum dimension.
These hypotheses address the concept of homogeneity of the test specimens.

Replication
The test was replicated four times. As each test takes a considerably long time, it was
not feasible to plan for replication more than four.

The Response Variable

After extensive group discussion, permanent deformation at 1000 load repetition,
p1000, was selected as the response variable. This is in line of previous experiments performed
in the field of pavement engineering.

Analysis procedures
The analysis procedure was determined in accordance with the standard procedure for
a 2K factorial design outlined in the text Design and Analysis of Experiments written by
Douglas C. Montgomery. The planned analysis steps are outlined in the next page.

1. Estimate factor effects

The group decided that the controlled factors should be modified as follows:
a) Nominal aggregate size (A): 12.5mm and 37.5mm
b) Specimen diameter (B): 70mm and 150 mm
c) Height to diameter ratio, H/D (C): 1 and 2.
2. Form initial model
The initial permanent deformation model would be:
PD = k0 + k1x1 + k2x2 + k3x3 + k4x1x2 + k5x1x3 + k6x2x3 +k7x1x2x3
Where x1, x2, x3 are the factors, k0 is the grand mean, kj
(j = 1, 2, ...7) is the effects and x1x2 , x1x3 , x2x3, x1x2x3 indicates interactions.
2. Perform statistical testing
Each controlled factor will be tested at two levels and each factor will be replicated
four times.
3. Refine model
If it is necessary, we will remove any insignificant effects from the permanent
deformation model.
4. Analysis
The following plot will be used for statistical analysis:
a) Normal probability plots of effects
b) Normal probability plots of residuals
c) Contour plots
5. Refine prediction model by proper transformation
6. Interpret results

Experimental Design
The final experimental design is shown below in Table 1.
Table 1 Summary of the Experimental Design
Standard
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32

Run
sequence
27
10
7
19
25
28
16
4
18
12
1
17
31
30
13
8
26
11
2
14
6
23
20
21
3
22
24
15
32
29
9
5

12.5
12.5
12.5
12.5
37.5
37.5
37.5
37.5
12.5
12.5
12.5
12.5
37.5
37.5
37.5
37.5
12.5
12.5
12.5
12.5
37.5
37.5
37.5
37.5
12.5
12.5
12.5
12.5
37.5
37.5
37.5
37.5

70
70
70
70
70
70
70
70
150
150
150
150
150
150
150
150
70
70
70
70
70
70
70
70
150
150
150
150
150
150
150
150

1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2

PD

Test Result
The test results are summarized in Table 2 below.
Table 2 Summary Test Results
Std
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32

Run
27
10
7
19
25
28
16
4
18
12
1
17
31
30
13
8
26
11
2
14
6
23
20
21
3
22
24
15
32
29
9
5

A
12.5
12.5
12.5
12.5
37.5
37.5
37.5
37.5
12.5
12.5
12.5
12.5
37.5
37.5
37.5
37.5
12.5
12.5
12.5
12.5
37.5
37.5
37.5
37.5
12.5
12.5
12.5
12.5
37.5
37.5
37.5
37.5

B
70
70
70
70
70
70
70
70
150
150
150
150
150
150
150
150
70
70
70
70
70
70
70
70
150
150
150
150
150
150
150
150

C
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2

PD
6850
6592
6218
7491
13427
9781
11780
10669
5334
5541
5673
5757
5888
4413
5524
8358
6234
6076
5663
6820
7109
10330
8826
8469
4760
5033
5101
4789
4190
6806
5498
5992

Analysis
Initial analysis
This experiment is a 23 design with 4 replicates. That means there are three factors (A,
B and C) each with two levels; one low and the other is high (as shown in the experimental
plan). The three factors that are of our interest are A(Nominal aggregate size), B(Specimen
diameter) and Hight to diameter (H/D). Table 3 shows the analysis of varience (ANOVA) that
summarizes the effect estimates and sum of squares. It is evident from the ANOVA that A, B,
C, and AB interaction are significant at about the 5 percent level.

Table 3 ANOVA in Initial Analysis

Response:
PD (Permanent Deformation)
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
F
Source
Squares
DF
Square
Value
Model
1.284E+008
7
1.834E+007 16.73
A
3.429E+007
1
3.429E+00731.26
B
5.963E+007
1
5.963E+00754.37
C
9.684E+006
1
9.684E+006
8.83
AB
1.766E+007
1
1.766E+007
16.10
AC
1.826E+006
1
1.826E+006
1.66
BC
2.508E+006
1
2.508E+006
2.29
ABC
2.812E+006
1
2.812E+006
2.56
Pure Error
2.632E+007
24
1.097E+006
Cor Total
1.547E+008
31
Std. Dev.
Mean
C.V.
PRESS

1047.22
6905.88
15.16
4.679E+007

Prob > F
< 0.0001 significant
< 0.0001
significant
< 0.0001
significant
0.0066
significant
0.0005
significant
0.2092
0.1436
0.1224

R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision 12.401

0.8299
0.7803
0.6976

Initial Predictive Equation

The initial predictive equation in terms of coded factors are shown below:
PD = 6905.88 + 1035.13 * A - 1365.06 * B - 550.13 * C - 742.81 * A * B
- 238.87 * A * C + 279.94 * B * C + 296.44 * A * B * C
The initial predictive equation in terms of actual factors are shown below:
PD = 2139.46875 + 499.2075 * A(NAS) + 26.48437 * B(SD) + 1576.40625*C(H/D)
-3.26425 *A(NAS) * B(SD) - 168.65250 *A (NAS) * C(H/D)
- 15.64687 * B(SD) * C(H/D) + 1.18575 * A(NAS) * B(SD) * C(H/D)

Half Normal Plot

A half normal plot (Figure 1, next page) was examined to find out significant factors.
From the plot it is evident that factor A, B, C and AB interaction are significant factors in this
experiment.

Model Refinement
As the half normal plot reveals that only A, B, C and AB interaction are significant, a
reanalysis was performed. Table 4 shows the revised ANOVA after removing the
nonsignificant factors.

H a lf N o r m a l p lo t
99

B
Half Normal % probability

97
95

A
AB

90

85
80
70
60
40
20
0

0 .0 0

6 8 2 .5 3

1 3 6 5 .0 6

2 0 4 7 .5 9

2 7 3 0 .1 3

|E ffe c t|

Figure 1 Half Normal Plot

Table 4 Revised ANOVA
Response:
PD
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
1.213E+008
4
3.031E+007
A
3.429E+007
1
3.429E+007
B
5.963E+007
1
5.963E+007
C
9.684E+006
1
9.684E+006
AB
1.766E+007
1
1.766E+007
Residual
3.347E+007
27
1.239E+006
Lack of Fit
7.146E+006
3
2.382E+006
Pure Error
2.632E+007
24
1.097E+006
Cor Total
1.547E+008
31
Std. Dev.
1113.32
R-Squared
Mean
6905.88
Adj R-Squared
C.V.
16.12
Pred R-Squared
PRESS4.701E+007

Adeq Precision

13.408

10

F
Value
24.46
27.66
48.11
7.81
14.25

Prob > F
< 0.0001
< 0.0001
< 0.0001
0.0094
0.0008

2.17

0.1176

0.7837
0.7517
0.6962

significant

not significant

Table 4 Revised ANOVA (Continued)

Factor
Intercept
A-A(NAS)
1.00
B-B(SD)
C-C(H/D)
AB

Coefficient
Estimate
6905.88
1035.13

DF
1
1

-1365.06
-550.13
-742.81

1
1
1

Standard
95% CI
Error
Low
196.81
6502.06
196.81 631.31
196.81-1768.88
196.81 -953.94
196.81-1146.63

95% CI
High
7309.69
-961.24
-146.31
-338.99

VIF

1.00
1.00
1.00

Revised Predictive Equation

The revised predictive equation in terms of coded factors are shown below:
PD = 6905.88 + 1035.13 * A - 1365.06 * B - 550.13 * C - 742.81 * A * B

The revised predictive equation in terms of actual factors are shown below:
PD = 6154.45313 + 246.22875 * A(NAS) +3.01406 * B(SD) - 1100.25 * C(H/D)
- 1.48562 * A(NAS) * B(SD)

Diagonostic Checking
Figure 2 is the normal probability plot of the residuals. There is clearly no ploblem
with normality. Figure 3 is the plot of the residuals versus the predicted values. Figure 4 is
the plot of residuals versus run number. It may be noticed that there are problems in model
fitting in Figure 3 and Figure 4. A data transformation is often used to deal with such
plroblems. So several data transformation were tried.

11

N o r m a l p lo t o f r e s id u a ls

Normal % probability

99
95
90
80
70
50
30
20
10
5
1

-2 . 3 4

-1 . 0 6

0 .2 1

1 .4 9

2 .7 7

S t u d e n t i z e d R e si d u a l s

Figure 2 Normal Probability Plot

R e s id u a ls v s . P r e d ic t e d

Studentized Residuals

3 .0 0

1 .5 0

0 .0 0

-1 . 5 0

-3 . 0 0

4 6 9 8 .3 8

6DESIGN-EX
1 7 3 .5 3
7 6 PERT
4 8 . 6 9 Plot
9 1 2 3 .8 4

1 0 5 9 9 .0 0

P re d i c t e d

Figure 3 Residual vs. Predicted Plot

PD

12

R e s id u a ls v s . R u n

Studentized Residuals

3 .0 0

1 .5 0

0 .0 0

-1 . 5 0

-3 . 0 0

11

16

21

26

31

Run Num ber

Figure 4 Residual vs. Run Number

Data Transformation
Figure 5 is the Box-Cox plot of power transformation. It was found from this plot that
the most apropriate data transformation is a inverse squre root.

Final ANOVA
The final ANOVA is shown in Table 5. Figure 6 is the normal probability plot of the
residuals. Figure 7 is the plot of the residuals versus the predicted values. Figure 8 is the plot
of residuals versus run number. There is no ploblem with these plots. Figure 9 is the plot of
predicted versus actual values of response variable. It shows good plot around the line of
equality, indeed. Figure 10 and Figure 11 represent one factor plot and a contour plot
respectively.

DESIGN-EXPERT Plot

13

B o x- C o x P lo t fo r P o w e r T r a n s fo r m s
1 8 .8 7

Ln(ResidualSS)

1 8 .4 0

1 7 .9 3

1 7 .4 6

1 6 .9 9

-3

-2

-1

Lam bda

Figure 5 Box-Cox Plot for Power Transforms

Table 5 Final ANOVA
Response:
PD
Transform: Inverse sqrt
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
7.012E-005
4
1.753E-005
A
1.631E-005
1
1.631E-005
B
4.211E-005
1
4.211E-005
C
5.113E-006
1
5.113E-006
AB
6.583E-006
1
6.583E-006
Residual
2.084E-005
27
7.720E-007
Lack of Fit
1.184E-006
3
3.948E-007
Pure Error
1.966E-005
24
8.191E-007
Cor Total
9.096E-005
31
Std. Dev.
8.786E-004
R-Squared
Mean
0.012
Adj R-Squared
C.V.
7.07
Pred R-Squared

14
DESIGN-EXPERT Plot

Constant:
F
Value
22.71
21.13
54.55
6.62
8.53
0.48
0.7709
0.7369
0.6781

Prob > F
< 0.0001 significant
< 0.0001
< 0.0001
0.0159
0.0070
0.6979

not significant

PRESS

2.928E-005

Adeq Precision

13.019

Table 5 Final ANOVA (Continued)

Factor
Intercept
A-A(NAS)
B-B(SD)
C-C(H/D)
AB

Coefficient
Estimate
0.012
-7.139E-004
1.147E-003
3.997E-004
4.536E-004

DF
1
1
1
1
1

Standard
95% CI
95% CI
Error
Low
High
1.553E-004
0.012
0.013
1.553E-004 -1.033E-003 -3.952E-004
1.553E-004 8.285E-004 1.466E-003
1.553E-004 8.105E-005 7.184E-004
1.553E-004 1.349E-004 7.723E-004

Final Equation in Terms of Coded Factors:

1.0/Sqrt(PD)
+0.012
-7.139E-004
+1.147E-003
+3.997E-004
+4.536E-004

=
*A
*B
*C
*A*B

Final Equation in Terms of Actual Factors:

1.0/Sqrt(PD)
=
+0.011989
-1.56901E-004
* A(NAS)
+6.00119E-006
* B(SD)
+7.99475E-004
* C(H/D)
+9.07147E-007 * A(NAS) * B(SD)

15

VIF
1.00
1.00
1.00
1.00

N o r m a l p lo t o f r e s id u a ls
99
95

Normal % probability

90
80
70
50
30
20
10
5
1

-2 .4 4

-1 .1 7

0 .1 1

1 .3 8

2 .6 6

S t u d e n ti z e d R e si d u a l s

Figure 6 Normal Probability Plot

R e s id u a ls v s . P r e d ic te d
3 .0 0

Studentized Residuals

1 .5 0

0 .0 0

-1 .5 0

-3 .0 0

0 .0 1 0

0 .0 1 1

0 .0 1 2

0 .0 1 3

0 .0 1 4

P re d i c te d

Figure 7 Residual vs. Predicted Plot

DESIGN-EXPERT Plot

16

R e s id u a ls v s . R u n
3 .0 0

Studentized Residuals

1 .5 0

0 .0 0

-1 . 5 0

-3 . 0 0

11

16

21

26

31

Run Num ber

Figure 8 Residual vs. Run Number

P r e d ic te d v s . A c tu a l
0 .0 1 5

Predicted

0 .0 1 4

0 .0 1 2

0 .0 1 0

0 .0 0 9

0 .0 0 9

0 .0 1 0

0 .0 1 2

0 .0 1 4

0 .0 1 5

A c tu a l

Figure 9 Residual vs. Actual Plot

DESIGN-EXPERT Plot

17

O n e F a c to r P lo t
13427

PD

1 1 1 1 7 .8

8 8 0 8 .5

6 4 9 9 .2 5

4190

1 .0 0

1 .2 5

1 .5 0

1 .7 5

2 .0 0

C : C (H / D )

Figure 10 One Factor Plot

1 5 0 .0 0

PD

B: B(SD)

1 3 0 .0 0

5 9 9 0 .0 6

1 1 0 .0 0

6 7 5 0 .4
9 0 .0 0

7 5 1 0 .7 4
8 2 7 1 .0 7

7 0 .0 0

1 2 .5 0

4
1 8 .7 5

2 5 .0 0

3 1 .2 5

3 7 .5 0

A : A (N A S )

Figure 11 Contour Plot for A vs. B

DESIGN-EXPERT Plot

18

Conclusion
From the analysis it has been found that the higher level of H/D ratio gives better result. The
contour plot reveals that low level of A and high level of B lead to a optimum result. The final
predictive equation is as follows:
1/Sqrt(PD) = 0.011989 -1.56901E-004 * A(NAS) + 6.00119E-006 * B(SD)
+7.99475E-004 * C(H/D) +9.07147E-007 * A(NAS) * B(SD)

19